MLA-C01 Question 303
Single answerYou are building a machine learning inference application hosted on EC2 Spot Instances behind an Auto Scaling group. The application experiences spikes in traffic during business hours, and maintaining low latency is critical. You need to decide between a target tracking scaling policy and a step scaling policy for the Auto Scaling group. Which scaling policy is most appropriate, and why?
- A
Use a target tracking scaling policy to automatically adjust the number of instances based on a defined metric, such as average CPU utilization, to maintain consistent performance.
- B
Use a step scaling policy to define specific scaling steps based on distinct metric thresholds, allowing fine-grained control over scaling behavior.
- C
Use a step scaling policy because it is designed to handle sudden traffic spikes more effectively than target tracking scaling.
- D
Use a target tracking scaling policy because it allows you to define multiple scaling steps for different traffic levels.
Show answer and explanation
Correct answer: A
Explanation
A target tracking scaling policy is better suited for applications with unpredictable traffic patterns and critical latency requirements. It dynamically adjusts the number of instances based on a target metric, such as average CPU utilization, ensuring consistent performance without requiring manual configuration of thresholds. In contrast, step scaling policies are more rigid and require predefined thresholds, making them less effective for highly variable traffic scenarios.
- A. Correct.
Correct: A target tracking scaling policy automatically adjusts the number of instances to maintain a target value for a defined metric, such as average CPU utilization. This is ideal for applications with variable traffic patterns and low-latency requirements, as it simplifies scaling decisions.
- B. Incorrect.
Incorrect: While step scaling policies allow fine-grained control, they are not as dynamic as target tracking policies for handling variable traffic patterns. They require pre-defined thresholds, which may not adapt well to unexpected traffic changes.
- C. Incorrect.
Incorrect: Step scaling policies are not inherently better at handling sudden traffic spikes compared to target tracking. Target tracking can respond dynamically to metric changes without manual thresholds.
- D. Incorrect.
Incorrect: Target tracking policies do not allow multiple scaling steps but instead focus on maintaining a target metric. This statement conflates the features of target tracking and step scaling policies.